Linpack is the "legacy" benchmark on pure FLOPs throughput. But HPCG is the more realistic benchmark that better tests the interconnects.
HPCG is a sparse-matrix, which more accurately represents many HPC problems. Linpack is a bit of a legacy, but still represents dense-compute pretty well.
Its still a GFLOPs / TFLops style measurement, but HPCG has a LOT more data movement.
I expect the Fujitsu ARM to be the greatest HPCG computer for years to come. No other design comes close.